Some JFR measurements showed that by default, on a large VM (1 TB) of a customer a huge amount of allocation outside TLAB occur.

Like a ~1hr JFR recording of only this event amounts to 2.3GB of data.

There might be issues like we had with PLAB sizing on G1 (e.g. JDK-8030849/JDK-8067341; maybe there are issues with region boundaries and something like a "net PLAB size" would be helpful).

Investigate and propose fixes.